Google acquired DeepMind for $400 million in 2014 and the firm became a subsidiary of Google’s parent company Alphabet in 2015.

Suleyman co-founded DeepMind and originally became its chief product officer. After Google’s acquisition in 2014, Suleyman became DeepMind’s head of applied AI.

In a tweet, Suleyman announced he’s now moving to Google itself to work on AI policy alongside Ken Walker, Jeff Dean, and others.

Suleyman is a longstanding proponent for AI ethics, so his presence at Google may help to ensure the web giant pushes boundaries in terms of technology but while respecting things like human rights.

Of course, DeepMind’s work hasn’t been without its controversies.

An app called Streams, developed by DeepMind for doctors and nurses, was ruled by the UK government in 2017 to have gained inappropriate access to medical data from 1.6 million patients.

At the time, Suleyman wrote in a blog post:

“DeepMind operates autonomously from Google, and we’ve been clear from the outset that at no stage will patient data ever be linked or associated with Google accounts, products or services.”

Critics believe this promise was broken when Streams was grabbed by Google itself. One thing which can’t be criticised, however, is DeepMind has shown its ability to get working AI solutions into the market. Suleyman appears to be part of Google’s desire to tap some of DeepMind’s ability to ship working AI solutions.

In a blog post, DeepMind co-founder Demis Hassabis spoke highly of the role Suleyman had played at the firm:

“As a serial entrepreneur, Mustafa played a key role over the past decade, helping to get DeepMind off the ground, and launched a series of innovative collaborations with Google to reduce energy consumption in data centres, improve Android battery performance, optimise Google Play, and find ways to improve the lives of patients, nurses and doctors alike,” he said.

“Mustafa leaves DeepMind having helped set us up for long-term success, and I’m looking forward to what he’ll achieve in the years ahead as he joins Google in a new role.”

Suleyman will be starting his new role at Google in January.

(Image Credit: Mustafa Suleyman by Joi Ito under CC BY 2.0 license)

Interested in hearing industry leaders discuss subjects like this? , , , AI &

The post DeepMind co-founder moves to Google to work on AI policy appeared first on AI News.

The consortium includes Google DeepMind, Facebook AI, New York University, and the University of Washington. Each of the consortium’s members believe a more comprehensive benchmark is needed for NLP than current solutions.

The result is a benchmarking platform called SuperGLUE which replaces an older platform called GLUE with a “much harder benchmark with comprehensive human baselines,” according to Facebook AI. 

SuperGLUE helps to put NLP abilities to the test where previous benchmarks were beginning to pose too simple for the latest systems.

“Within one year of release, several NLP models have already surpassed human baseline performance on the GLUE benchmark. Current models have advanced a surprisingly effective recipe that combines language model pretraining on huge text data sets with simple multitask and transfer learning techniques,” Facebook said.

In 2018, Google released BERT (Bidirectional Encoder Representations from Transformers) which Facebook calls one of the biggest breakthroughs in NLP. Facebook took Google’s open-source work and identified changes to improve its effectiveness which led to RoBERTa (Robustly Optimized BERT Pretraining Approach).

RoBERTa basically “smashed it,” as the kids would say, in commonly-used benchmarks:

“Within one year of release, several NLP models (including RoBERTa) have already surpassed human baseline performance on the GLUE benchmark. Current models have advanced a surprisingly effective recipe that combines language model pretraining on huge text data sets with simple multitask and transfer learning techniques,” Facebook explains.

For the SuperGLUE benchmark, the consortium decided on tasks which meet four criteria:

1. Have varied formats.
2. Use more nuanced questions.
3. Are yet-to-be-solved using state-of-the-art methods.
4. Can be easily solved by people.

The new benchmark includes eight diverse and challenging tasks, including a Choice of Plausible Alternatives (COPA) causal reasoning task. The aforementioned task provides the system with the premise of a sentence and it must determine either the cause or effect of the premise from two possible choices. Humans have managed to achieve 100 percent accuracy on COPA while BERT achieves just 74 percent.

Across SuperGLUE’s tasks, RoBERTa is currently the leading NLP system and isn’t far behind the human baseline:

You can find a full breakdown of SuperGLUE and its various benchmarking tasks in a Facebook AI blog post here.

deepgeniusai.com/">AI & Big Data Expo events with upcoming shows in Silicon Valley, London, and Amsterdam to learn more. Co-located with the IoT Tech Expo, , & .

The post Esteemed consortium launch AI natural language processing benchmark appeared first on AI News.

StarCraft 2 is a complex real-time strategy game that can still throw surprises at you even years after playing. In other words, StarCraft 2 is great for testing an AI.

Back in January, AI News reported DeepMind’s so-called ‘AlphaStar’ AI beat professional human eSports players Grzegorz Komincz and Dario Wunsch.

AlphaStar is now taking a virtual trip to Europe where it will begin playing a “small number” of games on the StarCraft 2 competitive ladder. Human players won’t even be aware they’re playing against the AI as it will be anonymised.

Blizzard explained the reasoning behind the anonymisation:

“Having AlphaStar play anonymously helps ensure that it is a controlled test so that the experimental versions of the agent experience gameplay as close to a normal 1v1 ladder match as possible.

It also helps ensure all games are played under the same conditions from match to match. DeepMind will release the research results in a peer-reviewed scientific paper along with replays of AlphaStar’s matches.”

AlphaStar was trained on historic game footage that StarCraft’s developer Blizzard has been releasing on a monthly basis. Five versions of the AI battled each other to hone their skills in training which equates to around 200 years for a human.

Multiple experimental variants of AlphaStar will take part in the test and it will play 1v1 matches only. The test will be opt-in, so players will have to click an in-game popup to get involved. Basically, it’s a voluntary ass-whooping.

deepgeniusai.com/">AI & Big Data Expo events with upcoming shows in Silicon Valley, London, and Amsterdam to learn more. Co-located with the IoT Tech Expo, , & .

The post Human-beating StarCraft 2 AI will compete anonymously in Europe appeared first on AI News.

The renowned Cambridge-based artificial intelligence experts hope their product will become parent company Alphabet’s first medical device.

In a live demonstration, a patient had a retinal scan performed on her eye. DeepMind’s system was able to provide a diagnosis and ‘urgency score’ in just 30 seconds.

The ability to perform such a quick diagnosis will free up doctors’ precious time for other purposes and ensure patients get treatment as soon as possible.

A range of eye diseases can already be diagnosed by DeepMind’s system including glaucoma, macular degeneration, and diabetic retinopathy.

Developed in conjunction with London’s Moorfields Eye Hospital, the system has the same level of accuracy as the world’s leading specialists.

Much like robotics and 5G are expected to help patients in rural areas gain access to world-leading experts, AI systems like DeepMind’s will provide the fastest and most accurate diagnosis of medical problems no matter where a patient is.

The product is being developed by DeepMind Health. The subsidiary has received its fair share of criticism after the UK government ruled the company had gained inappropriate access to medical data from 1.6 million patients when developing its Streams app.

Once boasting of a 100 percent employee retention rate, talent has begun leaking from DeepMind in recent months. Some believe the backlash from the Streams controversy has made it more difficult for DeepMind to find willing, and much-needed, partners in the medical industry.

DeepMind first published details about its research in science journal Nature last August.

deepgeniusai.com/">AI & Big Data Expo events with upcoming shows in Silicon Valley, London, and Amsterdam to learn more. Co-located with the IoT Tech Expo, , & .

The post DeepMind’s first commercial product diagnoses eye diseases appeared first on AI News.

Wind farms are a fantastic source of renewable energy but their output can be hard to predict due to the nature of wind.

By improving the accuracy of predictions and providing more data, wind farm operators are better prepared to meet upcoming power demands.

In a blog post, Google wrote:

“Based on these predictions, our model recommends how to make optimal hourly delivery commitments to the power grid a full day in advance. This is important, because energy sources that can be scheduled (i.e. can deliver a set amount of electricity at a set time) are often more valuable to the grid.”

Impressively, Google’s system is able to make accurate predictions up to 36 hours in advance:

Increased value helps to drive uptake of wind energy. While the algorithm is still being optimised, Google claims its AI has “boosted the value of our wind energy by roughly 20 percent”.

Here’s a comparison of using Google’s AI compared to a typical wind farm:

Google partnered with its subsidiary DeepMind to develop the AI which uses a neural network trained on weather forecasts and historical turbine data.

The project goes back somewhat to DeepMind’s roots at Google. One of DeepMind’s first achievements under Google was to reduce and control its vast data centres’ power consumption.

Google will rollout the AI optimisation across its central US wind farms which generate 700 megawatts of power.

deepgeniusai.com/">AI & Big Data Expo events with upcoming shows in Silicon Valley, London, and Amsterdam to learn more. Co-located with the IoT Tech Expo, , & .

The post Google is using AI to predict wind farm power output appeared first on AI News.

Back in 2015, DeepMind attempted to play various Atari games. The AI was competent in most of the games and became as good at Video Pinball as a human player.

DeepMind notoriously struggled with Montezuma’s Revenge, a notoriously complex game from the 1980s. The AI was unable to learn a path and retrieve even the first ‘key’ in the game.

Video games, in general, have become a battleground for AIs to show-off. DeepMind’s failure with Montezuma’s Revenge set the game as one benchmark for the industry to prove advancements.

A new algorithm designed by Fabio Zambetta and his team from RMIT University learns from past mistakes and identified next steps 10 times faster. The AI was successful in autonomously playing Montezuma’s Revenge.

In a statement, Zambetta explained:

“Truly intelligent AI needs to be able to learn to complete tasks autonomously in ambiguous environments.

We’ve shown that the right kind of algorithms can improve results using a smarter approach rather than purely brute forcing a problem end-to-end on very powerful computers.”

Zambetta presented his findings at the 33rd AAAI Conference on Artificial Intelligence in Hawaii today and explained how it works.

DeepMind and similar AIs struggle with adventure games like Montezuma’s Revenge due to a lack of reward until it obtains the first item, a key in this case. This makes it difficult for the AI to work out if what it’s doing is correct/optimal.

Games like Video Pinball provide the AI with quick rewards due to things such as point increases. This approach enables the AI to learn what path is going to achieve the highest score.

By implementing reinforcement learning, the researchers added ‘pellet rewards’ for the system to promote it exploring more paths.

“With time, this technology will be valuable to achieve goals in the real world, whether in self-driving cars or as useful robotic assistants with natural language recognition,” said Zambetta.

Other AI researchers are continuing to advance their approaches. DeepMind itself published two papers last summer describing how an AI could learn to conquer Montezuma’s Revenge from YouTube videos.

We look forward to watching upcoming AI bouts for the Montezuma’s Revenge title.

deepgeniusai.com/">AI & Big Data Expo events with upcoming shows in Silicon Valley, London, and Amsterdam to learn more. Co-located with the IoT Tech Expo, , & .

The post DeepMind’s AI bested in Atari game Montezuma’s Revenge appeared first on AI News.

The live stream of the showdowns was viewed by more than 55,000 people.

“This is, of course, an exciting moment for us,” said David Silver, a researcher at DeepMind. “For the first time, we saw an AI that was able to defeat a professional player.”

DeepMind created five versions of their ‘AlphaStar’ AI. Each AI was trained with historic game footage that StarCraft-developer Blizzard has been releasing on a monthly basis.

In order to further improve their abilities, the five AIs were pitted against each other in a league. The leading AI racked up experience that would equate to a human training for around 200 years.

Perhaps needless to say, AlphaStar wiped the floor with human players Grzegorz Komincz and Dario Wunsch.

You can watch AlphaStar taking on the human players below:

The only hope for humans so far is that AlphaStar was trained for a single map and using just the one alien race type of three available in the game. Removed from its comfort zone, it would not perform as well.

Video games have driven more rudimentary AI developments for decades. The advancement shown by AlphaStar could be used to create more complex ‘bots’ that can pose a challenge and help train even the best human players.

This isn’t the first time we’ve seen DeepMind’s AI bots in action – but, in the past, they’ve had a tendency of immediately rushing its opponents with ‘workers’ in a behaviour that Blizzard called “amusing”.

deepgeniusai.com/">AI & Big Data Expo events with upcoming shows in Silicon Valley, London, and Amsterdam to learn more. Co-located with the IoT Tech Expo, , & .

The post DeepMind thrashed pro StarCraft 2 players in latest demo appeared first on AI News.

StarCraft 2 is a complex game with many strategies, making it the perfect testing ground for AI. Google’s DeepMind first started exploring how it can use AI to beat the world’s best StarCraft players back in 2016.

In 2017, StarCraft’s developer Blizzard made 65,000 past matches available to DeepMind researchers to begin training bots. Blizzard promised it would make a further half a million games available each month.

We’ve seen DeepMind’s AI bots in action with various degrees of success. The AI had a tendency of immediately rushing its opponents with ‘workers’ in a behaviour that Blizzard called “amusing,” but confessed it had a 50 percent success rate even against StarCraft 2’s AI bots on ‘insane’ difficulty.

Fed with some replays from human players using more complex strategies, the AI began adopting them.

“After feeding the agent replays from real players, it started to execute standard macro-focused strategies, as well as defend against aggressive tactics such as cannon rushes,” Blizzard said.

We’re yet to see these new strategies being used by DeepMind’s AI but it won’t be much longer until we do.

“It’s only been a few months since BlizzCon but DeepMind is ready to share more information on their research,” Blizzard said today.

“The StarCraft games have emerged as a ‘grand challenge’ for the AI community as they’re the perfect environment for benchmarking progress against problems such as planning, dealing with uncertainty, and spatial reasoning.”

You can find a stream of DeepMind’s AI playing StarCraft 2 via StarCraft’s Twitch or Deepmind’s YouTube at 6pm GMT/10am PT/1pm ET on January 24th.

deepgeniusai.com/">AI & Big Data Expo events with upcoming shows in Silicon Valley, London, and Amsterdam to learn more. Co-located with the IoT Tech Expo, , & .

The post DeepMind’s AI will show off its new StarCraft 2 skills this week appeared first on AI News.

Speaking to The Guardian in 2016, DeepMind CEO Demis Hassabis said:

“We are able to literally get the best scientists from each country each year. So we’ll have, say, the person that won the Physics Olympiad in Poland, the person who got the top maths PhD of the year in France.

We’ve got more ideas than we’ve got researchers, but at the same time, there are more great people coming to our door than we can take on.”

In recent months, however, that renowned talent has started leaking out. Over the past 24 hours alone, two leading DeepMinders have announced they’re leaving the company.

Edward Grefenstette, a leading research scientist, announced he’s joining Facebook’s AI facility just down the road from DeepMind’s headquarters in London. Grefenstette co-founded Dark Blue Labs, a deep learning startup that DeepMind acquired in 2014.

Jack Kelly, a research engineer, has decided to leave DeepMind and launch a non-profit lab focused on fixing climate change. Kelly undertook the Computer Science MSc at Imperial College London with the explicit aim of using AI to mitigate climate change.

Such high-profile employees announcing their departure from a once 100 percent staff retention firm, within hours of each other, leads to wonder about deeper problems.

Outside the company, there are concerns about Google’s acquisition of DeepMind. These were reignited at the end of last year when Google absorbed DeepMind Health’s Streams app.

DeepMind Health was already controversial. In 2017, the UK government ruled the company had gained inappropriate access to medical data from 1.6 million patients when developing Streams.

Mustafa Suleyman, Head of Applied AI at DeepMind, wrote in a blog post:

“DeepMind operates autonomously from Google, and we’ve been clear from the outset that at no stage will patient data ever be linked or associated with Google accounts, products or services.”

Critics say this promise was broken when Streams was grabbed by Google.

The situation is sure to make potential future partners question whether to share data with DeepMind. Healthcare is one area set to benefit most from AI, yet it’s one that DeepMind could now struggle to find allies for.

Some of these concerns are sure to be held inside the company’s walls, but to what extent – and whether they’re resignation worthy – is unclear.

A pattern that has emerged from speaking to DeepMind employees is the mutual respect shared between researchers. At least one former employee maintains that DeepMind remains a great place to work.

deepgeniusai.com/">AI & Big Data Expo events with upcoming shows in Silicon Valley, London, and Amsterdam to learn more. Co-located with the IoT Tech Expo, , & .

The post Talent has begun leaking from DeepMind in recent months appeared first on AI News.

The AI, introduced by DeepMind in a research published in Science on 6th November 2018, was put against three of the world’s most complex board games and their current AI record holders.

This included world-champion chess AI Stockfish, DeepMind’s very own AlphaGo Zero – known as the best Go player in history – and elmo, winner of the 27th annual World Computer Shogi Championship in 2017.

AlphaZero defeated the AIs in all three games on its own and without human intervention. The only human assistance involved was teaching the AI the basic rules of the games. Using reinforcement learning, the AI would then play against itself millions of times using different strategies to win.

It took nine hours for the AI to learn chess, 12 hours for shogi, and 13 days for Go. Training the AI involving the use of a hefty 5,000 tensor processing units.

The learning algorithm was combined with the Monte Carlo tree search (MCTS), a ‘searching method’.  This is how the Go AI programmes know when and how to make the next move. This same system was used for chess and shogi, revealing the AI could be adapted to other complex games.

The interesting fact here is that AlphaZero implemented its own tactics which surprised human chess players. Moreover, the programme’s aggressive style and dynamic playstyle amazed chess grandmaster Matthew Sadler.

Such extraordinary abilities make the AI a fine teaching tool for chess players from which they can learn about hitherto-unseen gameplay strategies.

 AI & >.

The post AI beats AI: AlphaZero becomes the best game player in history appeared first on AI News.